Most business websites collect far more data than they use. Web analytics is the discipline that turns raw traffic into decisions: which channels deserve more budget, which pages quietly lose visitors, which change actually moved revenue. Done well, it takes a few hours to set up and thirty minutes a month to run. Done badly, it produces dashboards nobody opens. This guide covers the full practical chain: writing a measurement plan tied to business goals, choosing the right metrics, tracking events and conversions, disciplining campaign tagging, understanding what attribution can and cannot tell you, managing consent under GDPR, picking a tool category, building a dashboard people will actually read, and installing a monthly routine that ends in actions rather than screenshots.
What web analytics is, and what it is not?
Web analytics is the collection, measurement and interpretation of visitor behaviour on a site, with the explicit purpose of improving a business outcome. That last part is what most teams skip. Counting visits is measurement; deciding to rewrite a pricing page because seven out of ten visitors leave it without scrolling is analysis. The tool is only an instrument: the value comes from the questions you bring to it.
It is equally useful to be clear about what web analytics is not. It is not a complete census of your audience, not a substitute for talking to customers, and not a machine that settles internal debates on its own.
- Not exhaustive. Consent refusals, ad blockers and privacy-oriented browsers mean you always measure a sample, never the entire population.
- Not causal. Analytics shows what happened and in what order. Why requires a hypothesis, a test, or a conversation with a real user.
- Not qualitative. Numbers show where visitors drop off. Session recordings, surveys and support tickets explain the friction behind that drop-off.
- Not neutral. Every definition, what counts as a session, a conversion, a bounce, is a choice you make, and that choice shapes the answer.
Write a measurement plan before you choose a tool
The biggest mistake is installing a tracking script first and asking what to measure later. Reverse the order. A measurement plan is a short document that links business objectives to observable behaviour; it is what separates a useful installation from a data dump. It usually fits on one page.
From business objective to question, then to metric
Work in three steps. Start with the business objective, expressed in money or a countable outcome: forty qualified demo requests a month, a lower cost per lead, more repeat purchases. Then write the question that would tell you whether you are on track: which channels generate demo requests that become customers? Finally pick the metric that answers it. If a metric answers no question, and that question serves no objective, you do not need to track it.
The one-page measurement plan
- Objectives. Two or three at most, each with a numeric target and a timeframe.
- Primary conversions. The one or two actions that create real value: a qualified form, a paid order, a trial activation.
- Micro-conversions. The intermediate signals that predict conversion: a pricing page view, a document download, watching a video to the end.
- Segments. The cuts you will always look at: channel, device, country, new versus returning visitors.
- Owner and cadence. Who reads the data, how often, and what decision that person is authorised to take from it.
The metric families that matter in web analytics
Almost every useful metric belongs to one of five families. Thinking in families stops you stockpiling ten numbers that say the same thing, and makes blind spots obvious: if you track twelve acquisition metrics and none for retention, you already know where your gap is.
Acquisition: where visitors come from
Sessions and users by channel, source and campaign, with the new versus returning distinction. The useful version of this family is never a raw traffic total: it is traffic split by origin and judged on what it produces at the end of the journey. A channel that brings a tenth of your volume and a third of your revenue deserves attention no session counter would reveal.
Behaviour: what happens on the site
Landing pages, exit pages, pages per session, scroll depth, internal search terms and engaged time. Internal search is the most underused report in this family: it is a direct transcript of what visitors looked for and did not find. Read it every month and you get a free backlog of content to create and navigation problems to fix.
Conversion: the visitor actions that count
Conversion rate by channel and landing page, funnel step completion, form abandonment, and cost per acquisition if you run ads. Always express a conversion rate against an explicit denominator, per session, per user, or per visitor who reached a given step, and keep that definition stable, because changing it quietly makes every historical comparison unusable.
Retention and revenue: what happens next
Return rate, repeat purchase rate, subscription renewals, average order value and revenue per channel. This is the family small sites ignore most often, and yet it is the one that changes strategy most: a channel that acquires cheaply but whose customers leave quickly looks excellent in an acquisition report while destroying margin in reality.
Event and conversion tracking that stays maintainable
Modern tools are event-based: page views, clicks, form submissions and purchases are all events with parameters. That flexibility is a trap. Teams instrument every button on the site, end up with hundreds of event names, and six months later nobody knows whether the signup event fires on submit or on confirmation. Track deliberately.
Start with the conversions in your measurement plan, add the micro-conversions that precede them, then stop. Ten to twenty well-defined events cover the needs of most business websites. You can always add one later; cleaning a polluted event schema is far harder.
Naming conventions and documentation
- Name by object plus action. Names like form_submitted or checkout_completed stay readable; click2 or event_final do not.
- Fire on success, not intent. A form event triggered by a button click counts validation failures as conversions and inflates your numbers.
- Put variation in parameters. One form submission event with a parameter naming the form beats eight distinct event names.
- Document every event. A shared table listing name, trigger, parameters and owner takes twenty minutes and saves days.
- Test before you trust. Walk the real flow yourself in a debug view and confirm each event fires once, with the right values.
Campaign tagging and the limits of attribution
You cannot judge a channel you cannot identify, and you cannot fully credit a channel that works off-site. These two problems, tagging and attribution, are precisely where most reporting quietly falls apart.
UTM hygiene
UTM parameters are how you tell your tool where an inbound click came from. They are also entirely manual, so they degrade as soon as several people create links. Fix a convention and enforce it.
- Lowercase everything. Most tools treat Newsletter and newsletter as two different sources, which splits one channel into two rows.
- Freeze the vocabulary. Keep a short closed list of allowed values for source and medium, rather than inventing them per campaign.
- Use medium consistently. Medium describes the channel type, cpc, email, social, referral, and campaign names the specific operation.
- Never tag an internal link. A UTM between two of your own pages restarts the session and overwrites the original source.
- Build links from a single file. A shared spreadsheet or a link builder removes typos and leaves an auditable trail.
What attribution can and cannot tell you?
Attribution assigns credit for a conversion across the touchpoints that preceded it. Last-click gives everything to the final source, first-click to the discovery source, and data-driven models spread credit across the path. None is true in an absolute sense: they are filters, and each systematically flatters a different part of the funnel. Last-click overrates brand search and retargeting; first-click overrates awareness channels.
On top of the model choice come field constraints that shrink what you see. Cross-device journeys break when someone researches on mobile and buys on desktop. Cookie lifetimes truncate long consideration cycles. Consent refusals erase whole sessions from the path. Treat attribution reports as directional evidence, cross-check any major budget decision with a channel pause test or a simple self-reported question in your form, and distrust any decision that rests on a gap of a few percentage points.
Consent, privacy and the real effect of GDPR on your data
In the European Union, storing or reading non-essential cookies and similar identifiers requires informed consent collected before any tracking fires. This is a legal obligation, not an optimisation trade-off, and it has a direct consequence for your reports: a meaningful share of your visitors will never appear in your tool. Depending on your audience, sector and banner design, the share of lost sessions can range from a small minority to a large one, and consumer-facing sites typically see higher refusal rates than business-to-business ones.
- Block tracking until consent is given. Scripts must set no identifier before acceptance. Loading first and deleting later is not enough.
- Make refusal as easy as acceptance. A visible reject button on the first layer is both a regulatory expectation and a trust signal.
- Minimise what you collect. Never push email addresses, names or identifiers into analytics parameters or page URLs.
- Know your consent rate. Measure it and read every report knowing your data represents that share of reality.
- Compare like with like. A traffic rise after a banner redesign is almost always a change in consent rate, not in audience.
The practical conclusion is to stop chasing absolute accuracy. Consistent measurement is worth far more than complete measurement: if your consent rate is stable, trends and channel comparisons remain fully usable even though your totals are understated.
Choosing a web analytics tool category
Tools change constantly; categories do not. Decide first which category matches your questions, traffic volume and privacy posture, then choose a product inside it. Many teams end up combining two, usually a general platform plus a specialist.
- General platforms. Broad, free or low-cost, deeply integrated with advertising ecosystems. Relevant if you run paid ads and need channel and campaign tracking. Trade-off: complex interfaces and heavier consent obligations.
- Privacy-first tools. Lightweight, cookieless or nearly so, focused on aggregate traffic. Relevant for content sites and teams that want simple numbers with a smaller compliance surface. Trade-off: shallow funnels and limited individual analysis.
- Product analytics. Event-centric, built for funnels, cohorts and retention inside an application. Relevant once your site becomes a product with accounts and recurring use. Trade-off: cost and instrumentation effort.
- Server-side and warehouse tracking. Events emitted from your backend, often stored in your own database. Relevant for reliable revenue data and joining web behaviour to orders and customers. Trade-off: engineering time and ongoing maintenance.
- Qualitative tools. Heatmaps, session recordings, on-site surveys. They do not replace analytics, but they explain a drop-off the numbers have already flagged very quickly.
Building a dashboard that drives decisions
A dashboard is not a data dump. It is the answer to a recurring question, designed so anyone can tell in ten seconds whether the trajectory is healthy. If a chart has never changed a decision, delete it. That discipline of deletion is what lets a dashboard survive a busy week.
- One screen, five to eight numbers. Beyond that, people scroll without reading.
- Always show a comparison. A number without a previous period, a target or a benchmark carries no information.
- Order by the funnel. Sessions, then engagement, then conversions, then revenue: causes above effects.
- Segment the main line. A split by channel and one by device capture most of the variation worth explaining.
- Write the definitions on the dashboard. One line stating what counts as a conversion ends the recurring argument about whose number is right.
The classic web analytics mistakes
- Chasing vanity metrics. Page view totals and follower counts flatter the ego and change nothing. Prefer metrics you could act on if they moved.
- Tracking everything just in case. Unlimited events produce a schema nobody trusts. Track what your plan requires, then extend deliberately.
- Never defining the conversion. Without a named primary conversion, every performance debate becomes a matter of opinion.
- Neglecting data quality. Internal traffic, bots, duplicated tags and double-firing events silently corrupt reports for months.
- Reading a signal in small numbers. A conversion rate built on a handful of conversions swings wildly. Wait for volume or widen the period.
- Comparing incomparable periods. Seasonality, campaign spikes and tracking changes explain most surprising variations before any real behaviour change.
- Collecting with no owner. If nobody is accountable for looking at the numbers and acting, the installation will degrade within a quarter.
A monthly web analytics routine
Analytics creates value through regularity, not intensity. A focused monthly session of thirty to sixty minutes beats an occasional deep dive with no follow-through. Always use the same sequence so comparisons stay honest.
- 1. Check data quality first. Confirm your key events still fire, look for unexpected zeroes or spikes, and check that the consent rate has not moved.
- 2. Review the primary conversion. Volume and rate against last month and the same month last year.
- 3. Break it down by channel. Identify the channel that improved and the one that declined, and note the plausible cause.
- 4. Inspect the top landing pages. Look at the pages receiving the most entries and their conversion rates, not only their traffic.
- 5. Find one friction point. The largest drop from one step to the next in your main funnel is this month's project.
- 6. Write one decision. One change, one owner, one date. A month without a decision is a month of wasted measurement.
- 7. Close last month's loop. Check whether the previous decision produced the expected effect and record the answer.
How Cadrant helps you measure what you build?
Measurement becomes much simpler when the site and its data layer are designed together rather than reconciled afterwards. With Cadrant, you describe the site or application you need in natural language and get a production-ready result, backed by a Supabase database and published on your own domain. In practice, the conversions that matter are already recorded in a database you control, not only in a third-party report.
- Describe the actions that matter, a demo request, a booking, an order, and get structured records rather than anonymous hits.
- Join web behaviour to real business outcomes, because leads and orders live in your own Supabase tables.
- Publish on your own domain, which keeps referral data clean and avoids cross-domain attribution breaks.
- Add an internal dashboard page for your key numbers by describing it in a sentence, instead of rebuilding a spreadsheet every month.
- Iterate quickly on pages and funnels, so the improvements analytics identifies are actually live within the month.